@dxos/plugin-transcription 0.8.4-staging.60fe92afc8 → 0.9.1-main.c7dcc2e112

package/PLUGIN.mdl

@@ -69,6 +69,12 @@ type WhisperSegment
     words: WhisperWord[]
 ```
 
+```mdl
+type Settings
+  fields:
+    entityExtraction: boolean    # default: true — annotate transcript with references to known objects
+```
+
 ## Components
 
 ```mdl
@@ -169,6 +175,20 @@ op summarizeTranscript
     and clearly assigned.
 ```
 
+```mdl
+op enrichMessage
+  desc: Extracts proper nouns from each transcript block of a Message using a small LLM, full-text-searches the space for matching objects, and rewrites the text with inline `[noun](dxn)` reference links.
+  input:
+    message: Message
+  output:
+    message: Message
+  effects: [echo:read, ai:inference]
+  requires: [AiService, Database]
+  note: |
+    Uses Claude Haiku. Falls back to the raw message if extraction fails (e.g. the AI
+    service is unavailable). Gated by the entityExtraction setting.
+```
+
 ```mdl
 op normalizeSentences
   desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.

package/dist/lib/neutral/{TranscriptionArticle-L6XANBH5.mjs → TranscriptionArticle-S56VVOSF.mjs}

@@ -1,6 +1,6 @@
 import {
   renderByline
-} from "./chunk-KAEXRX3Y.mjs";
+} from "./chunk-XOXHWMXR.mjs";
 import "./chunk-J5LGTIGS.mjs";
 
 // src/containers/TranscriptionArticle/TranscriptionArticle.tsx
@@ -24,7 +24,7 @@ var TranscriptionArticle = ({ role, subject: transcript, attendableId }) => {
     label: [
       recording ? "stop-recording.label" : "start-recording.label",
       {
-        ns: meta.id
+        ns: meta.profile.key
       }
     ],
     icon: recording ? "ph--stop-circle--regular" : "ph--microphone--regular",
@@ -51,4 +51,4 @@ var TranscriptionArticle = ({ role, subject: transcript, attendableId }) => {
 export {
   TranscriptionArticle as default
 };
-//# sourceMappingURL=TranscriptionArticle-L6XANBH5.mjs.map
+//# sourceMappingURL=TranscriptionArticle-S56VVOSF.mjs.map

package/dist/lib/neutral/{TranscriptionArticle-L6XANBH5.mjs.map → TranscriptionArticle-S56VVOSF.mjs.map}

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/containers/TranscriptionArticle/TranscriptionArticle.tsx"],
-  "sourcesContent": ["//\n// Copyright 2024 DXOS.org\n//\n\nimport React from 'react';\n\nimport { type AppSurface } from '@dxos/app-toolkit/ui';\nimport { Filter, Obj, Query } from '@dxos/echo';\nimport { useMembers, useQuery } from '@dxos/react-client/echo';\nimport { Panel } from '@dxos/react-ui';\nimport { Menu, MenuBuilder, useMenuBuilder } from '@dxos/react-ui-menu';\nimport { Message, type Transcript } from '@dxos/types';\n\nimport { Transcription } from '#components';\nimport { useFeedModelAdapter, useTranscriptionRecording } from '#hooks';\nimport { meta } from '#meta';\n\nimport { renderByline } from '../../util';\n\nexport type TranscriptionArticleProps = AppSurface.ObjectArticleProps<Transcript.Transcript>;\n\nexport const TranscriptionArticle = ({ role, subject: transcript, attendableId }: TranscriptionArticleProps) => {\n  const db = Obj.getDatabase(transcript);\n  const members = useMembers(db?.spaceId).map((member) => member.identity);\n  const feed = transcript.feed.target;\n  const messages = useQuery(\n    db,\n    feed ? Query.select(Filter.type(Message.Message)).from(feed) : Query.select(Filter.nothing()),\n  );\n  const model = useFeedModelAdapter(renderByline(members), messages);\n  const { recording, toggleRecording } = useTranscriptionRecording(transcript);\n\n  const menuActions = useMenuBuilder(\n    () =>\n      MenuBuilder.make()\n        .action(\n          'toggle-recording',\n          {\n            label: [recording ? 'stop-recording.label' : 'start-recording.label', { ns: meta.id }],\n            icon: recording ? 'ph--stop-circle--regular' : 'ph--microphone--regular',\n            disposition: 'toolbar',\n            testId: 'transcription.toggle-recording',\n          },\n          toggleRecording,\n        )\n        .build(),\n    [recording, toggleRecording],\n  );\n\n  return (\n    <Panel.Root role={role}>\n      <Menu.Root {...menuActions} attendableId={attendableId}>\n        <Panel.Toolbar asChild>\n          <Menu.Toolbar />\n        </Panel.Toolbar>\n      </Menu.Root>\n      <Panel.Content asChild>\n        <Transcription model={model} transcript={transcript} />\n      </Panel.Content>\n    </Panel.Root>\n  );\n};\n"],
-  "mappings": ";;;;;;AAIA,OAAOA,WAAW;AAGlB,SAASC,QAAQC,KAAKC,aAAa;AACnC,SAASC,YAAYC,gBAAgB;AACrC,SAASC,aAAa;AACtB,SAASC,MAAMC,aAAaC,sBAAsB;AAClD,SAASC,eAAgC;AAEzC,SAASC,qBAAqB;AAC9B,SAASC,qBAAqBC,iCAAiC;AAC/D,SAASC,YAAY;AAMd,IAAMC,uBAAuB,CAAC,EAAEC,MAAMC,SAASC,YAAYC,aAAY,MAA6B;AACzG,QAAMC,KAAKC,IAAIC,YAAYJ,UAAAA;AAC3B,QAAMK,UAAUC,WAAWJ,IAAIK,OAAAA,EAASC,IAAI,CAACC,WAAWA,OAAOC,QAAQ;AACvE,QAAMC,OAAOX,WAAWW,KAAKC;AAC7B,QAAMC,WAAWC,SACfZ,IACAS,OAAOI,MAAMC,OAAOC,OAAOC,KAAKC,QAAQA,OAAO,CAAA,EAAGC,KAAKT,IAAAA,IAAQI,MAAMC,OAAOC,OAAOI,QAAO,CAAA,CAAA;AAE5F,QAAMC,QAAQC,oBAAoBC,aAAanB,OAAAA,GAAUQ,QAAAA;AACzD,QAAM,EAAEY,WAAWC,gBAAe,IAAKC,0BAA0B3B,UAAAA;AAEjE,QAAM4B,cAAcC,eAClB,MACEC,YAAYC,KAAI,EACbC,OACC,oBACA;IACEC,OAAO;MAACR,YAAY,yBAAyB;MAAyB;QAAES,IAAIC,KAAKC;MAAG;;IACpFC,MAAMZ,YAAY,6BAA6B;IAC/Ca,aAAa;IACbC,QAAQ;EACV,GACAb,eAAAA,EAEDc,MAAK,GACV;IAACf;IAAWC;GAAgB;AAG9B,SACE,sBAAA,cAACe,MAAMC,MAAI;IAAC5C;KACV,sBAAA,cAAC6C,KAAKD,MAAI;IAAE,GAAGd;IAAa3B;KAC1B,sBAAA,cAACwC,MAAMG,SAAO;IAACC,SAAAA;KACb,sBAAA,cAACF,KAAKC,SAAO,IAAA,CAAA,CAAA,GAGjB,sBAAA,cAACH,MAAMK,SAAO;IAACD,SAAAA;KACb,sBAAA,cAACE,eAAAA;IAAczB;IAActB;;AAIrC;",
-  "names": ["React", "Filter", "Obj", "Query", "useMembers", "useQuery", "Panel", "Menu", "MenuBuilder", "useMenuBuilder", "Message", "Transcription", "useFeedModelAdapter", "useTranscriptionRecording", "meta", "TranscriptionArticle", "role", "subject", "transcript", "attendableId", "db", "Obj", "getDatabase", "members", "useMembers", "spaceId", "map", "member", "identity", "feed", "target", "messages", "useQuery", "Query", "select", "Filter", "type", "Message", "from", "nothing", "model", "useFeedModelAdapter", "renderByline", "recording", "toggleRecording", "useTranscriptionRecording", "menuActions", "useMenuBuilder", "MenuBuilder", "make", "action", "label", "ns", "meta", "id", "icon", "disposition", "testId", "build", "Panel", "Root", "Menu", "Toolbar", "asChild", "Content", "Transcription"]
+  "sourcesContent": ["//\n// Copyright 2024 DXOS.org\n//\n\nimport React from 'react';\n\nimport { type AppSurface } from '@dxos/app-toolkit/ui';\nimport { Filter, Obj, Query } from '@dxos/echo';\nimport { useMembers, useQuery } from '@dxos/react-client/echo';\nimport { Panel } from '@dxos/react-ui';\nimport { Menu, MenuBuilder, useMenuBuilder } from '@dxos/react-ui-menu';\nimport { Message, type Transcript } from '@dxos/types';\n\nimport { Transcription } from '#components';\nimport { useFeedModelAdapter, useTranscriptionRecording } from '#hooks';\nimport { meta } from '#meta';\n\nimport { renderByline } from '../../util';\n\nexport type TranscriptionArticleProps = AppSurface.ObjectArticleProps<Transcript.Transcript>;\n\nexport const TranscriptionArticle = ({ role, subject: transcript, attendableId }: TranscriptionArticleProps) => {\n  const db = Obj.getDatabase(transcript);\n  const members = useMembers(db?.spaceId).map((member) => member.identity);\n  const feed = transcript.feed.target;\n  const messages = useQuery(\n    db,\n    feed ? Query.select(Filter.type(Message.Message)).from(feed) : Query.select(Filter.nothing()),\n  );\n  const model = useFeedModelAdapter(renderByline(members), messages);\n  const { recording, toggleRecording } = useTranscriptionRecording(transcript);\n\n  const menuActions = useMenuBuilder(\n    () =>\n      MenuBuilder.make()\n        .action(\n          'toggle-recording',\n          {\n            label: [recording ? 'stop-recording.label' : 'start-recording.label', { ns: meta.profile.key }],\n            icon: recording ? 'ph--stop-circle--regular' : 'ph--microphone--regular',\n            disposition: 'toolbar',\n            testId: 'transcription.toggle-recording',\n          },\n          toggleRecording,\n        )\n        .build(),\n    [recording, toggleRecording],\n  );\n\n  return (\n    <Panel.Root role={role}>\n      <Menu.Root {...menuActions} attendableId={attendableId}>\n        <Panel.Toolbar asChild>\n          <Menu.Toolbar />\n        </Panel.Toolbar>\n      </Menu.Root>\n      <Panel.Content asChild>\n        <Transcription model={model} transcript={transcript} />\n      </Panel.Content>\n    </Panel.Root>\n  );\n};\n"],
+  "mappings": ";;;;;;AAIA,OAAOA,WAAW;AAGlB,SAASC,QAAQC,KAAKC,aAAa;AACnC,SAASC,YAAYC,gBAAgB;AACrC,SAASC,aAAa;AACtB,SAASC,MAAMC,aAAaC,sBAAsB;AAClD,SAASC,eAAgC;AAEzC,SAASC,qBAAqB;AAC9B,SAASC,qBAAqBC,iCAAiC;AAC/D,SAASC,YAAY;AAMd,IAAMC,uBAAuB,CAAC,EAAEC,MAAMC,SAASC,YAAYC,aAAY,MAA6B;AACzG,QAAMC,KAAKC,IAAIC,YAAYJ,UAAAA;AAC3B,QAAMK,UAAUC,WAAWJ,IAAIK,OAAAA,EAASC,IAAI,CAACC,WAAWA,OAAOC,QAAQ;AACvE,QAAMC,OAAOX,WAAWW,KAAKC;AAC7B,QAAMC,WAAWC,SACfZ,IACAS,OAAOI,MAAMC,OAAOC,OAAOC,KAAKC,QAAQA,OAAO,CAAA,EAAGC,KAAKT,IAAAA,IAAQI,MAAMC,OAAOC,OAAOI,QAAO,CAAA,CAAA;AAE5F,QAAMC,QAAQC,oBAAoBC,aAAanB,OAAAA,GAAUQ,QAAAA;AACzD,QAAM,EAAEY,WAAWC,gBAAe,IAAKC,0BAA0B3B,UAAAA;AAEjE,QAAM4B,cAAcC,eAClB,MACEC,YAAYC,KAAI,EACbC,OACC,oBACA;IACEC,OAAO;MAACR,YAAY,yBAAyB;MAAyB;QAAES,IAAIC,KAAKC,QAAQC;MAAI;;IAC7FC,MAAMb,YAAY,6BAA6B;IAC/Cc,aAAa;IACbC,QAAQ;EACV,GACAd,eAAAA,EAEDe,MAAK,GACV;IAAChB;IAAWC;GAAgB;AAG9B,SACE,sBAAA,cAACgB,MAAMC,MAAI;IAAC7C;KACV,sBAAA,cAAC8C,KAAKD,MAAI;IAAE,GAAGf;IAAa3B;KAC1B,sBAAA,cAACyC,MAAMG,SAAO;IAACC,SAAAA;KACb,sBAAA,cAACF,KAAKC,SAAO,IAAA,CAAA,CAAA,GAGjB,sBAAA,cAACH,MAAMK,SAAO;IAACD,SAAAA;KACb,sBAAA,cAACE,eAAAA;IAAc1B;IAActB;;AAIrC;",
+  "names": ["React", "Filter", "Obj", "Query", "useMembers", "useQuery", "Panel", "Menu", "MenuBuilder", "useMenuBuilder", "Message", "Transcription", "useFeedModelAdapter", "useTranscriptionRecording", "meta", "TranscriptionArticle", "role", "subject", "transcript", "attendableId", "db", "Obj", "getDatabase", "members", "useMembers", "spaceId", "map", "member", "identity", "feed", "target", "messages", "useQuery", "Query", "select", "Filter", "type", "Message", "from", "nothing", "model", "useFeedModelAdapter", "renderByline", "recording", "toggleRecording", "useTranscriptionRecording", "menuActions", "useMenuBuilder", "MenuBuilder", "make", "action", "label", "ns", "meta", "profile", "key", "icon", "disposition", "testId", "build", "Panel", "Root", "Menu", "Toolbar", "asChild", "Content", "Transcription"]
 }

package/dist/lib/neutral/TranscriptionPlugin.mjs

@@ -4,12 +4,12 @@ import "./chunk-J5LGTIGS.mjs";
 import { Plugin } from "@dxos/app-framework";
 import { AppActivationEvents, AppPlugin } from "@dxos/app-toolkit";
 import { Transcript } from "@dxos/types";
-import { BlueprintDefinition, OperationHandler, ReactSurface, TextContent, Transcriber } from "#capabilities";
+import { BlueprintDefinition, OperationHandler, ReactSurface, TextContent, Transcriber, TranscriptionSettings } from "#capabilities";
 import { meta } from "#meta";
 import { translations } from "#translations";
 
-// raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl?raw
-var PLUGIN_default = '---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes blueprint-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: SerializationModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \xB7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   "latest utterance\u2026"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\u{1F3A4} Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript\'s feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript\'s ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer\'s ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the SerializationModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Blueprint Integration\n\n  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with "## <speaker> \xB7 <time>"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to "Stop Recording"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n';
+// raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl
+var PLUGIN_default = '---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes blueprint-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n```mdl\ntype Settings\n  fields:\n    entityExtraction: boolean    # default: true \u2014 annotate transcript with references to known objects\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: SerializationModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \xB7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   "latest utterance\u2026"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\u{1F3A4} Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript\'s feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop enrichMessage\n  desc: Extracts proper nouns from each transcript block of a Message using a small LLM, full-text-searches the space for matching objects, and rewrites the text with inline `[noun](dxn)` reference links.\n  input:\n    message: Message\n  output:\n    message: Message\n  effects: [echo:read, ai:inference]\n  requires: [AiService, Database]\n  note: |\n    Uses Claude Haiku. Falls back to the raw message if extraction fails (e.g. the AI\n    service is unavailable). Gated by the entityExtraction setting.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript\'s ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer\'s ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the SerializationModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Blueprint Integration\n\n  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with "## <speaker> \xB7 <time>"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to "Stop Recording"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n';
 
 // src/TranscriptionPlugin.tsx
 var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addBlueprintDefinitionModule({
@@ -26,13 +26,15 @@ var TranscriptionPlugin = Plugin.define(meta).pipe(AppPlugin.addBlueprintDefinit
   activate: ReactSurface
 }), AppPlugin.addTranslationsModule({
   translations
+}), AppPlugin.addSettingsModule({
+  activate: TranscriptionSettings
 }), Plugin.addModule({
   id: "transcription",
   activatesOn: AppActivationEvents.SetupAppGraph,
   activate: Transcriber
 }), AppPlugin.addPluginAssetModule({
   asset: {
-    pluginId: meta.id,
+    pluginId: meta.profile.key,
     path: "PLUGIN.mdl",
     content: PLUGIN_default,
     mimeType: "application/x-mdl"

package/dist/lib/neutral/TranscriptionPlugin.mjs.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
-  "sources": ["../../../src/TranscriptionPlugin.tsx", "raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl?raw"],
-  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\nimport { AppActivationEvents, AppPlugin } from '@dxos/app-toolkit';\nimport { Transcript } from '@dxos/types';\n\nimport { BlueprintDefinition, OperationHandler, ReactSurface, TextContent, Transcriber } from '#capabilities';\nimport { meta } from '#meta';\nimport { translations } from '#translations';\n\n// eslint-disable-next-line import/no-relative-packages\nimport pluginSpec from '../PLUGIN.mdl?raw';\n\nexport const TranscriptionPlugin = Plugin.define(meta).pipe(\n  AppPlugin.addBlueprintDefinitionModule({ activate: BlueprintDefinition }),\n  AppPlugin.addTextContentModule({ activate: TextContent }),\n  AppPlugin.addOperationHandlerModule({ activate: OperationHandler }),\n  AppPlugin.addSchemaModule({ schema: [Transcript.Transcript] }),\n  AppPlugin.addSurfaceModule({ activate: ReactSurface }),\n  AppPlugin.addTranslationsModule({ translations }),\n  Plugin.addModule({\n    id: 'transcription',\n    activatesOn: AppActivationEvents.SetupAppGraph,\n    activate: Transcriber,\n  }),\n  AppPlugin.addPluginAssetModule({\n    asset: { pluginId: meta.id, path: 'PLUGIN.mdl', content: pluginSpec, mimeType: 'application/x-mdl' },\n  }),\n  Plugin.make,\n);\n\nexport default TranscriptionPlugin;\n", "---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes blueprint-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: SerializationModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \u00B7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   \"latest utterance\u2026\"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\uD83C\uDFA4 Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript's feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript's ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer's ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the SerializationModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Blueprint Integration\n\n  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with \"## <speaker> \u00B7 <time>\"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to \"Stop Recording\"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n"],
-  "mappings": ";;;AAIA,SAASA,cAAc;AACvB,SAASC,qBAAqBC,iBAAiB;AAC/C,SAASC,kBAAkB;AAE3B,SAASC,qBAAqBC,kBAAkBC,cAAcC,aAAaC,mBAAmB;AAC9F,SAASC,YAAY;AACrB,SAASC,oBAAoB;;;ACV7B;;;ADeO,IAAMC,sBAAsBC,OAAOC,OAAOC,IAAAA,EAAMC,KACrDC,UAAUC,6BAA6B;EAAEC,UAAUC;AAAoB,CAAA,GACvEH,UAAUI,qBAAqB;EAAEF,UAAUG;AAAY,CAAA,GACvDL,UAAUM,0BAA0B;EAAEJ,UAAUK;AAAiB,CAAA,GACjEP,UAAUQ,gBAAgB;EAAEC,QAAQ;IAACC,WAAWA;;AAAY,CAAA,GAC5DV,UAAUW,iBAAiB;EAAET,UAAUU;AAAa,CAAA,GACpDZ,UAAUa,sBAAsB;EAAEC;AAAa,CAAA,GAC/ClB,OAAOmB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBC;EACjCjB,UAAUkB;AACZ,CAAA,GACApB,UAAUqB,qBAAqB;EAC7BC,OAAO;IAAEC,UAAUzB,KAAKkB;IAAIQ,MAAM;IAAcC,SAASC;IAAYC,UAAU;EAAoB;AACrG,CAAA,GACA/B,OAAOgC,IAAI;AAGb,IAAA,8BAAejC;",
-  "names": ["Plugin", "AppActivationEvents", "AppPlugin", "Transcript", "BlueprintDefinition", "OperationHandler", "ReactSurface", "TextContent", "Transcriber", "meta", "translations", "TranscriptionPlugin", "Plugin", "define", "meta", "pipe", "AppPlugin", "addBlueprintDefinitionModule", "activate", "BlueprintDefinition", "addTextContentModule", "TextContent", "addOperationHandlerModule", "OperationHandler", "addSchemaModule", "schema", "Transcript", "addSurfaceModule", "ReactSurface", "addTranslationsModule", "translations", "addModule", "id", "activatesOn", "AppActivationEvents", "SetupAppGraph", "Transcriber", "addPluginAssetModule", "asset", "pluginId", "path", "content", "pluginSpec", "mimeType", "make"]
+  "sources": ["../../../src/TranscriptionPlugin.tsx", "raw-loader:/__w/dxos/dxos/packages/plugins/plugin-transcription/PLUGIN.mdl"],
+  "sourcesContent": ["//\n// Copyright 2023 DXOS.org\n//\n\nimport { Plugin } from '@dxos/app-framework';\nimport { AppActivationEvents, AppPlugin } from '@dxos/app-toolkit';\nimport { Transcript } from '@dxos/types';\n\nimport {\n  BlueprintDefinition,\n  OperationHandler,\n  ReactSurface,\n  TextContent,\n  Transcriber,\n  TranscriptionSettings,\n} from '#capabilities';\nimport { meta } from '#meta';\nimport { translations } from '#translations';\n\n// eslint-disable-next-line import/no-relative-packages\nimport pluginSpec from '../PLUGIN.mdl?raw';\n\nexport const TranscriptionPlugin = Plugin.define(meta).pipe(\n  AppPlugin.addBlueprintDefinitionModule({ activate: BlueprintDefinition }),\n  AppPlugin.addTextContentModule({ activate: TextContent }),\n  AppPlugin.addOperationHandlerModule({ activate: OperationHandler }),\n  AppPlugin.addSchemaModule({ schema: [Transcript.Transcript] }),\n  AppPlugin.addSurfaceModule({ activate: ReactSurface }),\n  AppPlugin.addTranslationsModule({ translations }),\n  AppPlugin.addSettingsModule({ activate: TranscriptionSettings }),\n  Plugin.addModule({\n    id: 'transcription',\n    activatesOn: AppActivationEvents.SetupAppGraph,\n    activate: Transcriber,\n  }),\n  AppPlugin.addPluginAssetModule({\n    asset: { pluginId: meta.profile.key, path: 'PLUGIN.mdl', content: pluginSpec, mimeType: 'application/x-mdl' },\n  }),\n  Plugin.make,\n);\n\nexport default TranscriptionPlugin;\n", "---\nid: org.dxos.plugin.transcription\nname: TranscriptionPlugin\nversion: 0.1.0\n---\n\nA real-time voice-to-text transcription plugin for DXOS Composer.\nCaptures spoken audio from a microphone, converts it to timestamped text segments via the\nWhisper speech-recognition service, and stores the results in an ECHO `Transcript` object\nbacked by a `Feed`.\nThe plugin also exposes blueprint-compatible operations that let AI assistants open and\nsummarize transcripts inside agent workflows.\n\n## Extensions\n\nThe following extension dialects are used in this document.\nEach extension is defined in the Appendix or resolved via its URI.\n\n| Term        | URI                            |\n|-------------|--------------------------------|\n| `type`      | `org.dxos.mdl.type@1.0`       |\n| `feat`      | `org.dxos.mdl.feat@1.0`       |\n| `test`      | `org.dxos.mdl.test@1.0`       |\n| `component` | `org.dxos.mdl.component@1.0`  |\n| `op`        | `org.dxos.mdl.op@1.0`         |\n\n## Types\n\n```mdl\ntype TranscriptSegment\n  fields:\n    text: string              # transcribed text for this segment\n    started: string           # ISO-8601 wall-clock timestamp of segment start\n```\n\n```mdl\ntype Message\n  fields:\n    id: string\n    created: string           # ISO-8601 creation timestamp\n    blocks: TranscriptSegment[]\n    sender:\n      identityDid?: string    # DID of the speaker, if known\n```\n\n```mdl\ntype Transcript\n  fields:\n    id: string\n    feed: Ref<Feed>           # ECHO feed that stores Message objects\n    started?: string          # ISO-8601 timestamp when recording started\n```\n\n```mdl\ntype WhisperWord\n  fields:\n    word: string\n    start: number             # seconds from start of audio chunk\n    end: number\n```\n\n```mdl\ntype WhisperSegment\n  fields:\n    text: string\n    start: number\n    end: number\n    no_speech_prob: number    # probability of silence; used to filter noise\n    words: WhisperWord[]\n```\n\n```mdl\ntype Settings\n  fields:\n    entityExtraction: boolean    # default: true \u2014 annotate transcript with references to known objects\n```\n\n## Components\n\n```mdl\ncomponent Transcription\n  desc: Read-only CodeMirror editor that renders transcript messages as Markdown with byline headings.\n  props:\n    transcript?: Transcript\n    model: SerializationModel<Message>\n  slots:\n    (none \u2014 self-contained editor viewport)\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [CodeMirror editor \u2014 read-only] \u2502\n    \u2502  ## Speaker \u00B7 HH:MM             \u2502\n    \u2502  Transcribed text\u2026              \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent Oracle\n  desc: |\n    Ambient display surface with an SVG graph canvas and a centered text overlay.\n    Used to show the latest transcribed utterance alongside a live visualization.\n  props:\n    text?: string             # latest utterance to display\n    delay?: number            # per-character reveal delay in ms (default 12)\n  slots:\n    canvas?: ReactNode        # replaces default SVG.Graph when provided\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502  [SVG canvas \u2014 absolutely fills]\u2502\n    \u2502         [Oracle.Text]           \u2502\n    \u2502   \"latest utterance\u2026\"           \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n```mdl\ncomponent TranscriptionArticle\n  desc: Full-page article panel combining the Transcription editor with a start/stop recording toolbar action.\n  props:\n    subject: Transcript\n    role: string\n    attendableId?: string\n  state:\n    recording: boolean        # whether microphone capture is active\n  actions:\n    toggleRecording()\n  layout: |\n    \u250C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510\n    \u2502 [Toolbar]  [\uD83C\uDFA4 Start/Stop]      \u2502\n    \u251C\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2524\n    \u2502 [Transcription editor]          \u2502\n    \u2502  \u2026                              \u2502\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518\n```\n\n## Operations\n\nPure operations \u2014 no side-effects.\n\n```mdl\nop openTranscript\n  desc: Loads a Transcript from ECHO, reads all Messages from its Feed, and returns the full text content rendered with speaker bylines.\n  input:\n    transcript: Ref<Transcript>\n  output: string\n  errors:\n    TranscriptNotFound: no Transcript object found for the given ref\n    FeedNotFound: the Transcript's feed ref cannot be resolved\n  requires: [Database, FeedService]\n```\n\nEffectful operations \u2014 produce output via AI or mutate persistent state.\n\n```mdl\nop createTranscript\n  desc: Creates a new Feed and a Transcript object linked to it, then writes both into the current ECHO space.\n  input:\n    name?: string\n    space: Space\n  output: Transcript\n  effects: [echo:write]\n```\n\n```mdl\nop summarizeTranscript\n  desc: Sends a transcript string and optional participant notes to an LLM and returns a Markdown summary with key points, verbatim quotes, and actionable tasks extracted from the conversation.\n  input:\n    transcript: string\n    notes?: string\n  output: string\n  effects: [http]\n  requires: [AiService]\n  note: |\n    Uses Claude Sonnet. Prompt instructs the model to bold speaker names,\n    include verbatim quotes, and emit only tasks that are directly actionable\n    and clearly assigned.\n```\n\n```mdl\nop enrichMessage\n  desc: Extracts proper nouns from each transcript block of a Message using a small LLM, full-text-searches the space for matching objects, and rewrites the text with inline `[noun](dxn)` reference links.\n  input:\n    message: Message\n  output:\n    message: Message\n  effects: [echo:read, ai:inference]\n  requires: [AiService, Database]\n  note: |\n    Uses Claude Haiku. Falls back to the raw message if extraction fails (e.g. the AI\n    service is unavailable). Gated by the entityExtraction setting.\n```\n\n```mdl\nop normalizeSentences\n  desc: Post-processes an array of raw transcript messages, merging word-level tokens into grammatically complete sentences.\n  input:\n    messages: Message[]\n  output: Message[]\n  note: pure transformation; no effects or service dependencies\n```\n\n## Features\n\n```mdl\nfeat F-1: Recording Lifecycle\n\n  req F-1.1: User can start microphone capture from TranscriptionArticle toolbar.\n  req F-1.2: User can stop microphone capture; final buffered chunks are flushed and transcribed.\n  req F-1.3: TranscriptionManager restarts the Transcriber automatically when the audio track changes.\n```\n\n```mdl\nfeat F-2: Audio Chunking and Transcription\n\n  req F-2.1:\n    when: user is speaking and chunk count reaches transcribeAfterChunksAmount\n    then: Transcriber schedules a deferred transcription task\n\n  req F-2.2:\n    when: transcription task runs\n    then: buffered PCM chunks are merged into a WAV file and POSTed to the Whisper endpoint\n\n  req F-2.3:\n    when: Whisper response is received\n    then: segments are filtered by no_speech_prob and aligned to absolute wall-clock timestamps\n\n  req F-2.4:\n    when: user is not speaking\n    then: only prefixBufferChunksAmount most-recent chunks are retained; older chunks discarded\n```\n\n```mdl\nfeat F-3: Transcript Persistence\n\n  req F-3.1:\n    when: aligned segments are produced\n    then: a Message object with the segments is appended to the Transcript's ECHO Feed\n\n  req F-3.2:\n    when: a messageEnricher is configured\n    then: Message is passed through the enricher before being written to the Feed\n\n  req F-3.3:\n    when: a peer's ECHO space receives the Feed update\n    then: TranscriptionArticle re-renders with the new messages\n    tags: [collaborative]\n```\n\n```mdl\nfeat F-4: Transcript Display\n\n  req F-4.1:\n    when: a Transcript object is opened\n    then: all Messages are rendered in the Transcription editor with speaker bylines and timestamps\n\n  req F-4.2:\n    when: new messages arrive during live recording\n    then: the SerializationModel is updated and the editor scrolls to show the latest content\n```\n\n```mdl\nfeat F-5: AI Summary\n\n  req F-5.1:\n    when: op:summarizeTranscript is invoked\n    then: LLM returns a structured Markdown document with key points, quotes, and tasks\n\n  req F-5.2:\n    when: participant notes are provided\n    then: notes are correlated with transcript content in the summary\n\n  req F-5.3:\n    when: no actionable tasks are found\n    then: the tasks section is omitted from the summary\n```\n\n```mdl\nfeat F-6: Blueprint Integration\n\n  req F-6.1: Blueprint exposes op:openTranscript and op:summarizeTranscript as AI tools.\n  req F-6.2: An AI assistant can open a transcript by Ref and return formatted text to the user.\n  req F-6.3: An AI assistant can invoke the summary operation and present the result as a message.\n```\n\n## Acceptance\n\n```mdl\ntest T-1: Create transcript\n  given: a DXOS space is available\n  when: op:createTranscript is invoked\n  then:\n    - a new Feed is created in the space\n    - a Transcript object linked to the Feed is created in the space\n    - the returned Transcript.feed.target is not null\n```\n\n```mdl\ntest T-2: Record and persist segment\n  given: a Transcript with a Feed, microphone capture active\n  when: the user speaks and silence is detected\n  then:\n    - Whisper returns at least one segment\n    - a Message is appended to the Feed\n    - Message.blocks contains at least one TranscriptSegment with non-empty text\n```\n\n```mdl\ntest T-3: Deduplicate segments across chunk boundaries\n  given: two consecutive transcription tasks with overlapping Whisper output\n  when: the second task produces segments whose end timestamps fall before lastTimestamp\n  then:\n    - duplicate segments are filtered out\n    - only novel words (start >= lastTimestamp) are included\n```\n\n```mdl\ntest T-4: Open transcript\n  given: a Transcript with three Messages in its Feed\n  when: op:openTranscript is invoked\n  then:\n    - content string contains three byline sections\n    - each section starts with \"## <speaker> \u00B7 <time>\"\n```\n\n```mdl\ntest T-5: Summarize transcript\n  given: a transcript string and optional notes string\n  when: op:summarizeTranscript is invoked\n  then:\n    - summary is a non-empty Markdown string\n    - speaker names appear in bold\n    - a tasks section is present if actionable tasks were detected\n```\n\n```mdl\ntest T-6: Sentence normalization is pure\n  given: an array of raw word-level Messages\n  when: op:normalizeSentences is called\n  then:\n    - output messages contain complete sentences\n    - no ECHO writes occur\n    - input array is not mutated\n```\n\n```mdl\ntest T-7: Toggle recording in UI\n  given: TranscriptionArticle is mounted with a Transcript\n  when: user clicks the Start Recording toolbar button\n  then:\n    - recording state becomes true\n    - toolbar button label changes to \"Stop Recording\"\n    - microphone capture begins\n```\n\n---\n\n## Appendix: Extension Definitions\n\nExtension block types used in this document are defined below using\nthe core `ext` primitive \u2014 the only construct the base language provides.\n\n```mdl\next type\n  uri: org.dxos.mdl.type@1.0\n  desc: A named data structure with typed fields and optional literals.\n  fields:\n    desc?: Prose\n    fields?: FieldMap        # name[?]: TypeExpr  (# inline comment)\n    literals?: UnionList     # a | b | c\n    extends?: TypeRef[]\n```\n\n```mdl\next feat\n  uri: org.dxos.mdl.feat@1.0\n  desc: A named feature grouping one or more requirements.\n  fields:\n    desc?: Prose\n    req: RequirementList\n  nesting: self              # feat blocks may contain feat blocks\n```\n\n```mdl\next test\n  uri: org.dxos.mdl.test@1.0\n  desc: An acceptance scenario expressed as given / when / then steps.\n  fields:\n    given?: Step | Step[]\n    when?: Step | Step[]\n    then: Step | Step[]\n    tags?: TagList\n```\n\n```mdl\next component\n  uri: org.dxos.mdl.component@1.0\n  desc: A UI component with props, internal state, slots, actions, and events.\n  fields:\n    desc?: Prose\n    props?: FieldMap            # external inputs (immutable inside component)\n    state?: FieldMap            # internal reactive state\n    slots?: FieldMap            # named ReactNode injection points\n    actions?: ActionMap         # methods the component exposes or handles\n    emits?: EventMap            # events the component raises to its parent\n    layout?: CodeBlock          # ASCII sketch of visual structure (non-normative)\n```\n\n```mdl\next op\n  uri: org.dxos.mdl.op@1.0\n  desc: |\n    A named operation with typed inputs, outputs, and declared errors.\n    Pure ops have no effects or requires. Effectful ops declare both.\n  fields:\n    desc?: Prose\n    input?: FieldMap            # named input parameters\n    output?: TypeExpr           # return type\n    errors?: ErrorMap           # name: Prose  (when this error occurs)\n    effects?: EffectList        # echo:read | echo:write | http | fs | ...\n    requires?: ServiceList      # injected service dependencies\n    note?: Prose                # implementation guidance (non-normative)\n```\n"],
+  "mappings": ";;;AAIA,SAASA,cAAc;AACvB,SAASC,qBAAqBC,iBAAiB;AAC/C,SAASC,kBAAkB;AAE3B,SACEC,qBACAC,kBACAC,cACAC,aACAC,aACAC,6BACK;AACP,SAASC,YAAY;AACrB,SAASC,oBAAoB;;;ACjB7B;;;ADsBO,IAAMC,sBAAsBC,OAAOC,OAAOC,IAAAA,EAAMC,KACrDC,UAAUC,6BAA6B;EAAEC,UAAUC;AAAoB,CAAA,GACvEH,UAAUI,qBAAqB;EAAEF,UAAUG;AAAY,CAAA,GACvDL,UAAUM,0BAA0B;EAAEJ,UAAUK;AAAiB,CAAA,GACjEP,UAAUQ,gBAAgB;EAAEC,QAAQ;IAACC,WAAWA;;AAAY,CAAA,GAC5DV,UAAUW,iBAAiB;EAAET,UAAUU;AAAa,CAAA,GACpDZ,UAAUa,sBAAsB;EAAEC;AAAa,CAAA,GAC/Cd,UAAUe,kBAAkB;EAAEb,UAAUc;AAAsB,CAAA,GAC9DpB,OAAOqB,UAAU;EACfC,IAAI;EACJC,aAAaC,oBAAoBC;EACjCnB,UAAUoB;AACZ,CAAA,GACAtB,UAAUuB,qBAAqB;EAC7BC,OAAO;IAAEC,UAAU3B,KAAK4B,QAAQC;IAAKC,MAAM;IAAcC,SAASC;IAAYC,UAAU;EAAoB;AAC9G,CAAA,GACAnC,OAAOoC,IAAI;AAGb,IAAA,8BAAerC;",
+  "names": ["Plugin", "AppActivationEvents", "AppPlugin", "Transcript", "BlueprintDefinition", "OperationHandler", "ReactSurface", "TextContent", "Transcriber", "TranscriptionSettings", "meta", "translations", "TranscriptionPlugin", "Plugin", "define", "meta", "pipe", "AppPlugin", "addBlueprintDefinitionModule", "activate", "BlueprintDefinition", "addTextContentModule", "TextContent", "addOperationHandlerModule", "OperationHandler", "addSchemaModule", "schema", "Transcript", "addSurfaceModule", "ReactSurface", "addTranslationsModule", "translations", "addSettingsModule", "TranscriptionSettings", "addModule", "id", "activatesOn", "AppActivationEvents", "SetupAppGraph", "Transcriber", "addPluginAssetModule", "asset", "pluginId", "profile", "key", "path", "content", "pluginSpec", "mimeType", "make"]
 }

package/dist/lib/neutral/capabilities/index.mjs

@@ -3,15 +3,17 @@ import "../chunk-J5LGTIGS.mjs";
 // src/capabilities/index.ts
 import { Capability } from "@dxos/app-framework";
 var BlueprintDefinition = Capability.lazy("BlueprintDefinition", () => import("../blueprint-definition-DA4QFRYJ.mjs"));
-var TextContent = Capability.lazy("TextContent", () => import("../text-content-KBB37HSM.mjs"));
+var TextContent = Capability.lazy("TextContent", () => import("../text-content-XZ3HK4OT.mjs"));
 var OperationHandler = Capability.lazy("OperationHandler", () => import("../operation-handler-NXMOAXFP.mjs"));
 var ReactSurface = Capability.lazy("ReactSurface", () => import("../react-surface-YMZRO242.mjs"));
-var Transcriber = Capability.lazy("Transcriber", () => import("../transcriber-OGSF3XO5.mjs"));
+var Transcriber = Capability.lazy("Transcriber", () => import("../transcriber-KYV6NWI5.mjs"));
+var TranscriptionSettings = Capability.lazy("TranscriptionSettings", () => import("../settings-DUS32HOM.mjs"));
 export {
   BlueprintDefinition,
   OperationHandler,
   ReactSurface,
   TextContent,
-  Transcriber
+  Transcriber,
+  TranscriptionSettings
 };
 //# sourceMappingURL=index.mjs.map

package/dist/lib/neutral/capabilities/index.mjs.map

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../../src/capabilities/index.ts"],
-  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { Capability } from '@dxos/app-framework';\nimport { type AppCapabilities } from '@dxos/app-toolkit';\nimport type { OperationHandlerSet } from '@dxos/compute';\n\n// The contributed capability type references Blueprint types from @dxos/compute, so the lazy\n// wrapper needs an explicit annotation to keep the inferred type portable (TS2883).\nexport const BlueprintDefinition: Capability.LazyCapability<\n  void,\n  Capability.Capability<typeof AppCapabilities.BlueprintDefinition>[]\n> = Capability.lazy('BlueprintDefinition', () => import('./blueprint-definition'));\nexport const TextContent = Capability.lazy('TextContent', () => import('./text-content'));\nexport const OperationHandler = Capability.lazy<OperationHandlerSet.OperationHandlerSet>(\n  'OperationHandler',\n  () => import('./operation-handler'),\n);\nexport const ReactSurface = Capability.lazy('ReactSurface', () => import('./react-surface'));\nexport const Transcriber = Capability.lazy('Transcriber', () => import('./transcriber'));\n"],
-  "mappings": ";;;AAIA,SAASA,kBAAkB;AAMpB,IAAMC,sBAGTD,WAAWE,KAAK,uBAAuB,MAAM,OAAO,sCAAA,CAAA;AACjD,IAAMC,cAAcH,WAAWE,KAAK,eAAe,MAAM,OAAO,8BAAA,CAAA;AAChE,IAAME,mBAAmBJ,WAAWE,KACzC,oBACA,MAAM,OAAO,mCAAA,CAAA;AAER,IAAMG,eAAeL,WAAWE,KAAK,gBAAgB,MAAM,OAAO,+BAAA,CAAA;AAClE,IAAMI,cAAcN,WAAWE,KAAK,eAAe,MAAM,OAAO,6BAAA,CAAA;",
-  "names": ["Capability", "BlueprintDefinition", "lazy", "TextContent", "OperationHandler", "ReactSurface", "Transcriber"]
+  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { Capability } from '@dxos/app-framework';\nimport type { OperationHandlerSet } from '@dxos/compute';\n\nexport const BlueprintDefinition = Capability.lazy('BlueprintDefinition', () => import('./blueprint-definition'));\nexport const TextContent = Capability.lazy('TextContent', () => import('./text-content'));\nexport const OperationHandler = Capability.lazy<OperationHandlerSet.OperationHandlerSet>(\n  'OperationHandler',\n  () => import('./operation-handler'),\n);\nexport const ReactSurface = Capability.lazy('ReactSurface', () => import('./react-surface'));\nexport const Transcriber = Capability.lazy('Transcriber', () => import('./transcriber'));\nexport const TranscriptionSettings = Capability.lazy('TranscriptionSettings', () => import('./settings'));\n"],
+  "mappings": ";;;AAIA,SAASA,kBAAkB;AAGpB,IAAMC,sBAAsBD,WAAWE,KAAK,uBAAuB,MAAM,OAAO,sCAAA,CAAA;AAChF,IAAMC,cAAcH,WAAWE,KAAK,eAAe,MAAM,OAAO,8BAAA,CAAA;AAChE,IAAME,mBAAmBJ,WAAWE,KACzC,oBACA,MAAM,OAAO,mCAAA,CAAA;AAER,IAAMG,eAAeL,WAAWE,KAAK,gBAAgB,MAAM,OAAO,+BAAA,CAAA;AAClE,IAAMI,cAAcN,WAAWE,KAAK,eAAe,MAAM,OAAO,6BAAA,CAAA;AAChE,IAAMK,wBAAwBP,WAAWE,KAAK,yBAAyB,MAAM,OAAO,0BAAA,CAAA;",
+  "names": ["Capability", "BlueprintDefinition", "lazy", "TextContent", "OperationHandler", "ReactSurface", "Transcriber", "TranscriptionSettings"]
 }

package/dist/lib/neutral/{chunk-KMT3BMMQ.mjs → chunk-2URVUJT4.mjs}

@@ -2,21 +2,37 @@ import {
   __export
 } from "./chunk-J5LGTIGS.mjs";
 
+// src/types/Settings.ts
+var Settings_exports = {};
+__export(Settings_exports, {
+  Settings: () => Settings
+});
+import * as Schema from "effect/Schema";
+var Settings = Schema.mutable(Schema.Struct({
+  entityExtraction: Schema.optional(Schema.Boolean.annotations({
+    title: "Entity extraction",
+    description: "While transcribing, use the Assistant to detect and annotate mentions of known objects such as people or organizations."
+  })).pipe(Schema.withConstructorDefault(() => true))
+}));
+
 // src/types/TranscriptionCapabilities.ts
 var TranscriptionCapabilities_exports = {};
 __export(TranscriptionCapabilities_exports, {
+  Settings: () => Settings2,
   TranscriberProvider: () => TranscriberProvider,
   TranscriptionManagerProvider: () => TranscriptionManagerProvider
 });
 import { Capability } from "@dxos/app-framework";
 import { meta } from "#meta";
-var TranscriberProvider = Capability.make(`${meta.id}.capability.transcriber`);
-var TranscriptionManagerProvider = Capability.make(`${meta.id}.capability.transcription-manager`);
+var TranscriberProvider = Capability.make(`${meta.profile.key}.capability.transcriber`);
+var TranscriptionManagerProvider = Capability.make(`${meta.profile.key}.capability.transcription-manager`);
+var Settings2 = Capability.make(`${meta.profile.key}.capability.settings`);
 
 // src/types/TranscriptOperation.ts
 var TranscriptOperation_exports = {};
 __export(TranscriptOperation_exports, {
   Create: () => Create,
+  EnrichMessage: () => EnrichMessage,
   MessageWithRangeId: () => MessageWithRangeId,
   Open: () => Open,
   SentenceNormalization: () => SentenceNormalization,
@@ -24,30 +40,30 @@ __export(TranscriptOperation_exports, {
   SentenceNormalizationOutput: () => SentenceNormalizationOutput,
   Summarize: () => Summarize
 });
-import * as Schema from "effect/Schema";
+import * as Schema2 from "effect/Schema";
 import { AiService } from "@dxos/ai";
 import { SpaceSchema } from "@dxos/client/echo";
 import { Operation } from "@dxos/compute";
 import { Database, Feed, Ref, Type, DXN } from "@dxos/echo";
 import { Message, Transcript } from "@dxos/types";
 import { meta as meta2 } from "#meta";
-var makeKey = (name) => DXN.make(`${meta2.id}.operation.${name}`);
+var makeKey = (name) => DXN.make(`${meta2.profile.key}.operation.${name}`);
 var Create = Operation.make({
   meta: {
     key: makeKey("create"),
     name: "Create Transcript",
     icon: "ph--microphone--regular"
   },
-  input: Schema.Struct({
-    name: Schema.optional(Schema.String),
+  input: Schema2.Struct({
+    name: Schema2.optional(Schema2.String),
     space: SpaceSchema
   }),
-  output: Schema.Struct({
+  output: Schema2.Struct({
     object: Type.getSchema(Transcript.Transcript)
   })
 });
-var MessageWithRangeId = Schema.extend(Type.getSchema(Message.Message), Schema.Struct({
-  rangeId: Schema.optional(Schema.Array(Schema.String)).annotations({
+var MessageWithRangeId = Schema2.extend(Type.getSchema(Message.Message), Schema2.Struct({
+  rangeId: Schema2.optional(Schema2.Array(Schema2.String)).annotations({
     description: "The IDs of the messages that contain the sentences."
   })
 }));
@@ -58,13 +74,13 @@ var Open = Operation.make({
     description: "Opens and reads the contents of a transcription object.",
     icon: "ph--folder-open--regular"
   },
-  input: Schema.Struct({
+  input: Schema2.Struct({
     transcript: Ref.Ref(Transcript.Transcript).annotations({
       description: "The ID of the transcription object."
     })
   }),
-  output: Schema.Struct({
-    content: Schema.String
+  output: Schema2.Struct({
+    content: Schema2.String
   }),
   services: [
     Database.Service,
@@ -78,16 +94,16 @@ var Summarize = Operation.make({
     description: "Summarize a transcript of a meeting.",
     icon: "ph--text-align-left--regular"
   },
-  input: Schema.Struct({
-    transcript: Schema.String.annotations({
+  input: Schema2.Struct({
+    transcript: Schema2.String.annotations({
       description: "The transcript of the meeting."
     }),
-    notes: Schema.optional(Schema.String).annotations({
+    notes: Schema2.optional(Schema2.String).annotations({
       description: "Additional notes from the participants."
     })
   }),
-  output: Schema.Struct({
-    summary: Schema.String.annotations({
+  output: Schema2.Struct({
+    summary: Schema2.String.annotations({
       description: "The summary of the transcript."
     })
   }),
@@ -95,13 +111,31 @@ var Summarize = Operation.make({
     AiService.AiService
   ]
 });
-var SentenceNormalizationInput = Schema.Struct({
-  messages: Schema.Array(MessageWithRangeId).annotations({
+var EnrichMessage = Operation.make({
+  meta: {
+    key: DXN.make("org.dxos.function.transcription.enrichMessage"),
+    name: "Enrich Transcript Message",
+    description: "Extract proper nouns from a transcript message and link them to objects in the space.",
+    icon: "ph--text-t--regular"
+  },
+  input: Schema2.Struct({
+    message: Type.getSchema(Message.Message)
+  }),
+  output: Schema2.Struct({
+    message: Type.getSchema(Message.Message)
+  }),
+  services: [
+    AiService.AiService,
+    Database.Service
+  ]
+});
+var SentenceNormalizationInput = Schema2.Struct({
+  messages: Schema2.Array(MessageWithRangeId).annotations({
     description: "Messages to normalize into sentences."
   })
 });
-var SentenceNormalizationOutput = Schema.Struct({
-  sentences: Schema.Array(MessageWithRangeId.pipe(Schema.mutable)).pipe(Schema.mutable).annotations({
+var SentenceNormalizationOutput = Schema2.Struct({
+  sentences: Schema2.Array(MessageWithRangeId.pipe(Schema2.mutable)).pipe(Schema2.mutable).annotations({
     description: "The sentences of the transcript."
   })
 });
@@ -118,7 +152,8 @@ var SentenceNormalization = Operation.make({
 });
 
 export {
+  Settings_exports,
   TranscriptionCapabilities_exports,
   TranscriptOperation_exports
 };
-//# sourceMappingURL=chunk-KMT3BMMQ.mjs.map
+//# sourceMappingURL=chunk-2URVUJT4.mjs.map

package/dist/lib/neutral/chunk-2URVUJT4.mjs.map

@@ -0,0 +1,7 @@
+{
+  "version": 3,
+  "sources": ["../../../src/types/Settings.ts", "../../../src/types/TranscriptionCapabilities.ts", "../../../src/types/TranscriptOperation.ts"],
+  "sourcesContent": ["//\n// Copyright 2026 DXOS.org\n//\n\n// @import-as-namespace\n\nimport * as Schema from 'effect/Schema';\n\nexport const Settings = Schema.mutable(\n  Schema.Struct({\n    entityExtraction: Schema.optional(\n      Schema.Boolean.annotations({\n        title: 'Entity extraction',\n        description:\n          'While transcribing, use the Assistant to detect and annotate mentions of known objects such as people or organizations.',\n      }),\n    ).pipe(Schema.withConstructorDefault(() => true)),\n  }),\n);\n\nexport interface Settings extends Schema.Schema.Type<typeof Settings> {}\n", "//\n// Copyright 2025 DXOS.org\n//\n\n// @import-as-namespace\n\nimport { type Atom } from '@effect-atom/atom-react';\n\nimport { Capability } from '@dxos/app-framework';\n\nimport { meta } from '#meta';\n\nimport {\n  type MediaStreamRecorderProps,\n  type Transcriber as TranscriberType,\n  type TranscriberProps,\n  type TranscriptMessageEnricher,\n  type TranscriptionManager as TranscriptionManagerType,\n} from '../transcriber';\nimport * as Settings$ from './Settings'; // eslint-disable-line\n\nexport type TranscriberProviderProps = {\n  audioStreamTrack: MediaStreamTrack;\n  recorderConfig?: Partial<MediaStreamRecorderProps['config']>;\n  transcriberConfig?: Partial<TranscriberProps['config']>;\n  onSegments: TranscriberProps['onSegments'];\n  transcribe?: TranscriberProps['transcribe'];\n};\n\nexport type TranscriberProvider = (props: TranscriberProviderProps) => TranscriberType;\n\nexport type TranscriptionManagerProviderProps = {\n  messageEnricher?: TranscriptMessageEnricher;\n};\n\nexport type TranscriptionManagerProvider = (props: TranscriptionManagerProviderProps) => TranscriptionManagerType;\n\nexport const TranscriberProvider = Capability.make<TranscriberProvider>(`${meta.profile.key}.capability.transcriber`);\n\nexport const TranscriptionManagerProvider = Capability.make<TranscriptionManagerProvider>(\n  `${meta.profile.key}.capability.transcription-manager`,\n);\n\nexport const Settings = Capability.make<Atom.Writable<Settings$.Settings>>(`${meta.profile.key}.capability.settings`);\n", "//\n// Copyright 2025 DXOS.org\n//\n\n// @import-as-namespace\n\nimport * as Schema from 'effect/Schema';\n\nimport { AiService } from '@dxos/ai';\nimport { SpaceSchema } from '@dxos/client/echo';\nimport { Operation } from '@dxos/compute';\nimport { Database, Feed, Ref, Type, DXN } from '@dxos/echo';\nimport { Message, Transcript } from '@dxos/types';\n\nimport { meta } from '#meta';\n\nconst makeKey = (name: string) => DXN.make(`${meta.profile.key}.operation.${name}`);\n\nexport const Create = Operation.make({\n  meta: { key: makeKey('create'), name: 'Create Transcript', icon: 'ph--microphone--regular' },\n  input: Schema.Struct({\n    name: Schema.optional(Schema.String),\n    space: SpaceSchema,\n  }),\n  output: Schema.Struct({\n    object: Type.getSchema(Transcript.Transcript),\n  }),\n});\n\nexport const MessageWithRangeId = Schema.extend(\n  Type.getSchema(Message.Message),\n  Schema.Struct({\n    rangeId: Schema.optional(Schema.Array(Schema.String)).annotations({\n      description: 'The IDs of the messages that contain the sentences.',\n    }),\n  }),\n);\n\nexport type MessageWithRangeIdType = Schema.Schema.Type<typeof MessageWithRangeId>;\n\nexport const Open = Operation.make({\n  meta: {\n    key: DXN.make('org.dxos.function.transcription.open'),\n    name: 'Open',\n    description: 'Opens and reads the contents of a transcription object.',\n    icon: 'ph--folder-open--regular',\n  },\n  input: Schema.Struct({\n    transcript: Ref.Ref(Transcript.Transcript).annotations({\n      description: 'The ID of the transcription object.',\n    }),\n  }),\n  output: Schema.Struct({\n    content: Schema.String,\n  }),\n  services: [Database.Service, Feed.FeedService],\n});\n\nexport const Summarize = Operation.make({\n  meta: {\n    key: DXN.make('org.dxos.function.transcription.summarize'),\n    name: 'Summarize',\n    description: 'Summarize a transcript of a meeting.',\n    icon: 'ph--text-align-left--regular',\n  },\n  input: Schema.Struct({\n    transcript: Schema.String.annotations({\n      description: 'The transcript of the meeting.',\n    }),\n    notes: Schema.optional(Schema.String).annotations({\n      description: 'Additional notes from the participants.',\n    }),\n  }),\n  output: Schema.Struct({\n    summary: Schema.String.annotations({\n      description: 'The summary of the transcript.',\n    }),\n  }),\n  services: [AiService.AiService],\n});\n\nexport const EnrichMessage = Operation.make({\n  meta: {\n    key: DXN.make('org.dxos.function.transcription.enrichMessage'),\n    name: 'Enrich Transcript Message',\n    description: 'Extract proper nouns from a transcript message and link them to objects in the space.',\n    icon: 'ph--text-t--regular',\n  },\n  input: Schema.Struct({\n    message: Type.getSchema(Message.Message),\n  }),\n  output: Schema.Struct({\n    message: Type.getSchema(Message.Message),\n  }),\n  services: [AiService.AiService, Database.Service],\n});\n\nexport const SentenceNormalizationInput = Schema.Struct({\n  messages: Schema.Array(MessageWithRangeId).annotations({\n    description: 'Messages to normalize into sentences.',\n  }),\n});\n\nexport type SentenceNormalizationInputType = Schema.Schema.Type<typeof SentenceNormalizationInput>;\n\nexport const SentenceNormalizationOutput = Schema.Struct({\n  sentences: Schema.Array(MessageWithRangeId.pipe(Schema.mutable)).pipe(Schema.mutable).annotations({\n    description: 'The sentences of the transcript.',\n  }),\n});\n\nexport const SentenceNormalization = Operation.make({\n  meta: {\n    key: DXN.make('org.dxos.function.transcription.sentenceNormalization'),\n    name: 'Sentence Normalization',\n    description: 'Post process of transcription for sentence normalization',\n    icon: 'ph--text-t--regular',\n  },\n  input: SentenceNormalizationInput,\n  output: SentenceNormalizationOutput,\n  services: [],\n});\n"],
+  "mappings": ";;;;;AAAA;;;;AAMA,YAAYA,YAAY;AAEjB,IAAMC,WAAkBC,eACtBC,cAAO;EACZC,kBAAyBC,gBAChBC,eAAQC,YAAY;IACzBC,OAAO;IACPC,aACE;EACJ,CAAA,CAAA,EACAC,KAAYC,8BAAuB,MAAM,IAAA,CAAA;AAC7C,CAAA,CAAA;;;ACjBF;;kBAAAC;EAAA;;;AAQA,SAASC,kBAAkB;AAE3B,SAASC,YAAY;AA2Bd,IAAMC,sBAAsBF,WAAWG,KAA0B,GAAGF,KAAKG,QAAQC,GAAG,yBAAyB;AAE7G,IAAMC,+BAA+BN,WAAWG,KACrD,GAAGF,KAAKG,QAAQC,GAAG,mCAAmC;AAGjD,IAAMN,YAAWC,WAAWG,KAAwC,GAAGF,KAAKG,QAAQC,GAAG,sBAAsB;;;AC3CpH;;;;;;;;;;;AAMA,YAAYE,aAAY;AAExB,SAASC,iBAAiB;AAC1B,SAASC,mBAAmB;AAC5B,SAASC,iBAAiB;AAC1B,SAASC,UAAUC,MAAMC,KAAKC,MAAMC,WAAW;AAC/C,SAASC,SAASC,kBAAkB;AAEpC,SAASC,QAAAA,aAAY;AAErB,IAAMC,UAAU,CAACC,SAAiBL,IAAIM,KAAK,GAAGH,MAAKI,QAAQC,GAAG,cAAcH,IAAAA,EAAM;AAE3E,IAAMI,SAASd,UAAUW,KAAK;EACnCH,MAAM;IAAEK,KAAKJ,QAAQ,QAAA;IAAWC,MAAM;IAAqBK,MAAM;EAA0B;EAC3FC,OAAcC,eAAO;IACnBP,MAAaQ,iBAAgBC,cAAM;IACnCC,OAAOrB;EACT,CAAA;EACAsB,QAAeJ,eAAO;IACpBK,QAAQlB,KAAKmB,UAAUhB,WAAWA,UAAU;EAC9C,CAAA;AACF,CAAA;AAEO,IAAMiB,qBAA4BC,eACvCrB,KAAKmB,UAAUjB,QAAQA,OAAO,GACvBW,eAAO;EACZS,SAAgBR,iBAAgBS,cAAaR,cAAM,CAAA,EAAGS,YAAY;IAChEC,aAAa;EACf,CAAA;AACF,CAAA,CAAA;AAKK,IAAMC,OAAO9B,UAAUW,KAAK;EACjCH,MAAM;IACJK,KAAKR,IAAIM,KAAK,sCAAA;IACdD,MAAM;IACNmB,aAAa;IACbd,MAAM;EACR;EACAC,OAAcC,eAAO;IACnBc,YAAY5B,IAAIA,IAAII,WAAWA,UAAU,EAAEqB,YAAY;MACrDC,aAAa;IACf,CAAA;EACF,CAAA;EACAR,QAAeJ,eAAO;IACpBe,SAAgBb;EAClB,CAAA;EACAc,UAAU;IAAChC,SAASiC;IAAShC,KAAKiC;;AACpC,CAAA;AAEO,IAAMC,YAAYpC,UAAUW,KAAK;EACtCH,MAAM;IACJK,KAAKR,IAAIM,KAAK,2CAAA;IACdD,MAAM;IACNmB,aAAa;IACbd,MAAM;EACR;EACAC,OAAcC,eAAO;IACnBc,YAAmBZ,eAAOS,YAAY;MACpCC,aAAa;IACf,CAAA;IACAQ,OAAcnB,iBAAgBC,cAAM,EAAES,YAAY;MAChDC,aAAa;IACf,CAAA;EACF,CAAA;EACAR,QAAeJ,eAAO;IACpBqB,SAAgBnB,eAAOS,YAAY;MACjCC,aAAa;IACf,CAAA;EACF,CAAA;EACAI,UAAU;IAACnC,UAAUA;;AACvB,CAAA;AAEO,IAAMyC,gBAAgBvC,UAAUW,KAAK;EAC1CH,MAAM;IACJK,KAAKR,IAAIM,KAAK,+CAAA;IACdD,MAAM;IACNmB,aAAa;IACbd,MAAM;EACR;EACAC,OAAcC,eAAO;IACnBuB,SAASpC,KAAKmB,UAAUjB,QAAQA,OAAO;EACzC,CAAA;EACAe,QAAeJ,eAAO;IACpBuB,SAASpC,KAAKmB,UAAUjB,QAAQA,OAAO;EACzC,CAAA;EACA2B,UAAU;IAACnC,UAAUA;IAAWG,SAASiC;;AAC3C,CAAA;AAEO,IAAMO,6BAAoCxB,eAAO;EACtDyB,UAAiBf,cAAMH,kBAAAA,EAAoBI,YAAY;IACrDC,aAAa;EACf,CAAA;AACF,CAAA;AAIO,IAAMc,8BAAqC1B,eAAO;EACvD2B,WAAkBjB,cAAMH,mBAAmBqB,KAAYC,eAAO,CAAA,EAAGD,KAAYC,eAAO,EAAElB,YAAY;IAChGC,aAAa;EACf,CAAA;AACF,CAAA;AAEO,IAAMkB,wBAAwB/C,UAAUW,KAAK;EAClDH,MAAM;IACJK,KAAKR,IAAIM,KAAK,uDAAA;IACdD,MAAM;IACNmB,aAAa;IACbd,MAAM;EACR;EACAC,OAAOyB;EACPpB,QAAQsB;EACRV,UAAU,CAAA;AACZ,CAAA;",
+  "names": ["Schema", "Settings", "mutable", "Struct", "entityExtraction", "optional", "Boolean", "annotations", "title", "description", "pipe", "withConstructorDefault", "Settings", "Capability", "meta", "TranscriberProvider", "make", "profile", "key", "TranscriptionManagerProvider", "Schema", "AiService", "SpaceSchema", "Operation", "Database", "Feed", "Ref", "Type", "DXN", "Message", "Transcript", "meta", "makeKey", "name", "make", "profile", "key", "Create", "icon", "input", "Struct", "optional", "String", "space", "output", "object", "getSchema", "MessageWithRangeId", "extend", "rangeId", "Array", "annotations", "description", "Open", "transcript", "content", "services", "Service", "FeedService", "Summarize", "notes", "summary", "EnrichMessage", "message", "SentenceNormalizationInput", "messages", "SentenceNormalizationOutput", "sentences", "pipe", "mutable", "SentenceNormalization"]
+}

package/dist/lib/neutral/chunk-J4ONEYPJ.mjs

@@ -0,0 +1,8 @@
+// src/operations/index.ts
+import { OperationHandlerSet } from "@dxos/compute";
+var TranscriptionOperationHandlerSet = OperationHandlerSet.lazy(() => import("./create-7NGOP7JP.mjs"), () => import("./open-42H72C4G.mjs"), () => import("./summarize-45WVUPED.mjs"), () => import("./enrich-message-F2AOM53F.mjs"));
+
+export {
+  TranscriptionOperationHandlerSet
+};
+//# sourceMappingURL=chunk-J4ONEYPJ.mjs.map

package/dist/lib/neutral/{chunk-KXJZEYS7.mjs.map → chunk-J4ONEYPJ.mjs.map}

@@ -1,7 +1,7 @@
 {
   "version": 3,
   "sources": ["../../../src/operations/index.ts"],
-  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { OperationHandlerSet } from '@dxos/compute';\n\nexport const TranscriptionOperationHandlerSet = OperationHandlerSet.lazy(\n  () => import('./create'),\n  () => import('./open'),\n  () => import('./summarize'),\n);\n"],
-  "mappings": ";AAIA,SAASA,2BAA2B;AAE7B,IAAMC,mCAAmCD,oBAAoBE,KAClE,MAAM,OAAO,uBAAA,GACb,MAAM,OAAO,qBAAA,GACb,MAAM,OAAO,0BAAA,CAAA;",
+  "sourcesContent": ["//\n// Copyright 2025 DXOS.org\n//\n\nimport { OperationHandlerSet } from '@dxos/compute';\n\nexport const TranscriptionOperationHandlerSet = OperationHandlerSet.lazy(\n  () => import('./create'),\n  () => import('./open'),\n  () => import('./summarize'),\n  () => import('./enrich-message'),\n);\n"],
+  "mappings": ";AAIA,SAASA,2BAA2B;AAE7B,IAAMC,mCAAmCD,oBAAoBE,KAClE,MAAM,OAAO,uBAAA,GACb,MAAM,OAAO,qBAAA,GACb,MAAM,OAAO,0BAAA,GACb,MAAM,OAAO,+BAAA,CAAA;",
   "names": ["OperationHandlerSet", "TranscriptionOperationHandlerSet", "lazy"]
 }

package/dist/lib/neutral/{chunk-TO77KAQK.mjs → chunk-XMHH34NI.mjs}

@@ -168,6 +168,7 @@ var useTranscriber = ({ audioStreamTrack, recorderConfig, transcriberConfig, onS
 // src/hooks/useTranscriptionRecording.ts
 import * as Effect from "effect/Effect";
 import { useCallback, useEffect as useEffect5, useState as useState3 } from "react";
+import { useAtomCapability, useOperationInvoker } from "@dxos/app-framework/ui";
 import { createFeedServiceLayer } from "@dxos/client/echo";
 import { Feed, Obj } from "@dxos/echo";
 import { EffectEx } from "@dxos/effect";
@@ -175,6 +176,7 @@ import { log as log2 } from "@dxos/log";
 import { getSpace } from "@dxos/react-client/echo";
 import { useIdentity } from "@dxos/react-client/halo";
 import { Message } from "@dxos/types";
+import { TranscriptOperation, TranscriptionCapabilities as TranscriptionCapabilities2 } from "#types";
 var __dxlog_file2 = "/__w/dxos/dxos/packages/plugins/plugin-transcription/src/hooks/useTranscriptionRecording.ts";
 var useTranscriptionRecording = (transcript) => {
   const space = getSpace(transcript);
@@ -182,6 +184,8 @@ var useTranscriptionRecording = (transcript) => {
   const feed = transcript.feed.target;
   const [recording, setRecording] = useState3(false);
   const track = useAudioTrack(recording);
+  const { invokePromise } = useOperationInvoker();
+  const settings = useAtomCapability(TranscriptionCapabilities2.Settings);
   const handleSegments = useCallback(async (blocks) => {
     if (!space || !feed || blocks.length === 0) {
       return;
@@ -190,19 +194,35 @@ var useTranscriptionRecording = (transcript) => {
       identityDid: identity.did,
       name: identity.profile?.displayName
     } : {};
-    const message = Obj.make(Message.Message, {
+    let message = Obj.make(Message.Message, {
       sender,
       created: (/* @__PURE__ */ new Date()).toISOString(),
       blocks
     });
-    const feedServiceLayer = createFeedServiceLayer(space.queues);
+    if (settings?.entityExtraction ?? true) {
+      const { data, error } = await invokePromise(TranscriptOperation.EnrichMessage, {
+        message
+      }, {
+        spaceId: space.id
+      });
+      if (error) {
+        log2.warn("entity extraction failed; using raw transcript", {
+          error
+        }, { "~LogMeta": "~LogMeta", F: __dxlog_file2, L: 53, S: void 0 });
+      } else if (data?.message) {
+        message = data.message;
+      }
+    }
+    const feedServiceLayer = createFeedServiceLayer(space.db);
     await Feed.append(feed, [
       message
     ]).pipe(Effect.provide(feedServiceLayer), EffectEx.runAndForwardErrors);
   }, [
     space,
     feed,
-    identity
+    identity,
+    invokePromise,
+    settings
   ]);
   const transcriber = useTranscriber({
     audioStreamTrack: track,
@@ -219,7 +239,7 @@ var useTranscriptionRecording = (transcript) => {
       }
     }, (err) => {
       if (!cancelled) {
-        log2.catch(err, void 0, { "~LogMeta": "~LogMeta", F: __dxlog_file2, L: 67, S: void 0 });
+        log2.catch(err, void 0, { "~LogMeta": "~LogMeta", F: __dxlog_file2, L: 90, S: void 0 });
         setRecording(false);
       }
     });
@@ -295,4 +315,4 @@ export {
   useTranscriptionRecording,
   useVoiceInput
 };
-//# sourceMappingURL=chunk-TO77KAQK.mjs.map
+//# sourceMappingURL=chunk-XMHH34NI.mjs.map